Why 'items' isn't 'patients', and why nobody tells you.

There is a moment, usually about slide four of any market-access deck, when someone says the word 'patients' while pointing at a chart of prescribing items. It's an easy slip. It's also often wrong by ten to forty percent, depending on the molecule.

What NHSBSA actually publishes

The English Prescribing Data (EPD) release counts items: a single line on a single prescription form. A patient on twice-daily therapy for a chronic condition may generate twelve items a year. A patient on a five-day acute course generates one.

It helps to know what one row of EPD actually contains, because it is less than most people assume. You get the dispensing month. You get the practice that issued the prescription, with its geography attached. You get the presentation, which is the drug, strength and form, coded against the BNF hierarchy and the NHS dictionary of medicines and devices. You get the number of items, the total quantity dispensed, the net ingredient cost and the actual cost. That is the whole row.

Notice what is missing. No NHS number. No age, no sex, no diagnosis. No way of telling whether the 26 packs of apixaban a practice dispensed last month went to 26 people or to 13. EPD is an accounting record, built so that dispensing reimbursement can be checked, and it is very good at that job. It was never designed to count people, and it does not.

Two of those columns get confused constantly, so it is worth being precise. Items is the number of prescription lines written and dispensed. Quantity is the number of tablets, millilitres or doses those lines added up to. One item for 56 tablets and one item for 28 tablets are both exactly one item. If you only ever look at items, a quiet change in prescription-duration policy will look like a market event. We will come back to that.

The gap, in numbers

For apixaban across England in the year to May 2026, primary-care items sit at about 17.1m. Applied to the average dose and pack size, that corresponds to roughly 780k unique patients on therapy at any point during the year. The ratio isn't stable. It changes with new-patient starts, adherence, and dose changes.

Here is the arithmetic, so you can check it rather than trust it. The standard maintenance dose of apixaban is 5mg twice daily, which is two tablets a day. A 28-tablet pack therefore lasts 14 days, and a patient who takes it every single day of the year needs 26 packs. If each prescription item covers one pack, a perfectly adherent, never-titrated, never-hospitalised patient generates 26 items a year.

No real cohort behaves like that. People start therapy in August rather than January, so their first year is a part year. People die, switch to another anticoagulant, or simply stop collecting. Some run gaps between packs. Some prescriptions cover two packs rather than one, which halves the item count for identical therapy. Fold all of that in and the observed average lands nearer 22 items per patient per year than the theoretical 26. Divide 17.1m items by 22 and you get roughly 780,000 people who received apixaban at some point in the year.

There is a second route to a related number, and it is worth running as a cross-check. Take a single month instead of a year. May 2026 shows about 1.42m primary-care items. A patient actively on 14-day packs collects around 2.2 items a month, so the point-in-time population is a little over 650,000 people on therapy that month. The annual figure sits higher than the monthly one because it also counts everyone who started, stopped or died part-way through. If your two routes disagree wildly, one of your assumptions is broken.

Every number in that chain is an assumption you can argue with, which is precisely the point. Move the adherence discount from 85% to 75% and the estimate shifts by tens of thousands of patients. Assume a larger share of 56-tablet packs and it shifts again. A patient estimate published without its assumptions attached is not an estimate. It is a decoration.

Where the ratio breaks

The 22-items figure is specific to apixaban, and it does not travel. An acute antibiotic like amoxicillin runs close to one item per patient per course, so items and patients are nearly the same number. A twice-daily chronic therapy runs at twenty-plus. A portfolio slide that applies one conversion factor across an acute product and a chronic one misstates both, in opposite directions.

Dose titration breaks it in a different way. A patient being stabilised on levothyroxine may hold 25, 50 and 100 microgram strengths at the same time, and each strength is its own item on the same prescription form. Warfarin patients routinely hold two or three tablet strengths so the week's dose can be assembled. One patient, three items, every month. Apixaban has a milder version of the same problem: the 2.5mg dose reduction means some patients generate items at two strengths in the same year as renal function or age moves them between doses.

Then there is behaviour with no clinical content at all. Items rise every December as patients collect early ahead of the holidays, and the rise unwinds in January. March 2020 remains the extreme case: a month of pandemic stockpiling that still distorts year-on-year comparisons if you forget it is sitting in the base. Quietest of all are prescription-duration policies. An ICB that moves from 28-day to 56-day prescribing halves its item count overnight with not one patient fewer, and ICBs genuinely differ on this. Neither policy is wrong. Both are just people managing supply. But compare two ICBs on items without checking duration policy and you are comparing their prescribing committees, not their populations.

A checklist for honest patient estimates

When we publish a patient estimate, it has been through five questions first. None of them needs anything fancier than a spreadsheet, and all of them need writing down. We would suggest yours should be too.

First, start from quantity rather than items wherever the data allows. Tablets dispensed, divided by tablets per patient per day, is a sturdier bridge than items divided by an assumed items-per-patient figure, because quantity is immune to pack size and duration policy.

Second, write the dosing assumption down. For apixaban we assume two tablets a day across the mix of 5mg and 2.5mg dosing, and we say so wherever the estimate appears. If the dose mix shifts, the estimate shifts with it, and a reader deserves to see that dependency rather than discover it.

Third, apply an adherence and persistence discount, and state it. Published adherence for oral anticoagulants sits broadly in the 80 to 90 percent range. Pick a figure, name it, and show what the estimate looks like at either end of the range. For once-daily therapies the discount matters less. For twice-daily dosing it compounds.

Fourth, decide what 'a patient' means before you compute anything. Treated at any point during the year is one definition. On therapy in a given month is another, and for apixaban it is 15 to 20 percent smaller, because of all the starts, stops and deaths in between. Both definitions are legitimate. Mixing them in one deck is not.

Fifth, sense-check the answer against something independent of prescribing data. Condition prevalence, QOF registers, hospital admissions. If atrial fibrillation prevalence supports about 1.6m diagnosed patients in England and your arithmetic puts 3m of them on a single anticoagulant, the arithmetic lost.

When items are the better metric

After all that, a short defence of the humble item. If the question is about dispensing workload, supply resilience or budget impact, items and quantity are exactly the right units, because pharmacies, wholesalers and medicines budgets deal in packs rather than people. Nobody has ever reimbursed a patient-year. A finance team asking about next year's drugs bill wants items and costs, not an epidemiology estimate.

Items are also hard to beat for direction of travel. So long as pack sizes and duration policies hold still, a 12.4% rise in items is a 12.4% rise in dispensed demand, whoever that demand belongs to, because the conversion factor cancels out of the comparison. That is why our monthly briefings lead with items, label them as items, and reach for a patient estimate only when the question genuinely needs people in it.

What to do about it

If your slide needs a patient number, say so and derive it. If it needs an items number, use items and be honest with the label. Anywhere DoseTrend shows one, we tell you which.

Written by the Cooply Solutions team. Corrections welcome at hello [at] dosetrend [dot] com.